diff --git a/CHANGELOG.md b/CHANGELOG.md index 79d7473b..b9a2189d 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -7,6 +7,30 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0 ## [Unreleased] +### Changed + +- **Claim-accuracy sweep across the public docs** (pre-v0.11.0; every change + traced to the 2026-07-15 primary-source verification pass): CLI↔GUI parity + claims recomputed from the manifest (98% → **93.4%**, 99 full / 7 api-only / + 0 cli-only — README badge, `parity-coverage.md` regenerated, positioning, + enterprise-grade, ROADMAP); count drift fixed to the machine-gated truths + (96 catalogs / 16 crosswalks / 14 collectors / 13 MCP tools / 4,900+ tests); + the FedRAMP first-mover registry row updated to the shipped `conmon ksi` + verb + CR26 SDR schema framing with explicit caveats, and two new + first-mover rows registered (M-25-21 practice/waiver-clock tracking, + M-25-22 acquisition lifecycle); the CISA SbD Pledge described as an + alignment self-attestation everywhere (never a signatory claim; formal + signing planned at v1.0); OSCAL-MCP landscape statements aligned to "one + of several" (CISO Assistant `ca_mcp` + trestle-mcp exist); the stale + "Sept 30 2026 FedRAMP mandate" framing replaced with CR26 FRC-CSO-JSN + (mandatory 2027-01-01) + NTC-0009 (2027-11-01, Class D High, + format-agnostic — never an "OSCAL mandate"); the unverifiable "18 months + ahead" claim removed; GovReady-Q described as dormant, not defunct; + OpenVEX / VSA rows moved to planned-tense with no version commitment; the + federal-SI walk-through refreshed (M-25-21 primary lens consistently, + M-26-05 rescission of the M-22-18 self-attestation form, the v0.11 + federal-wave surfaces added, resolved limitations pruned). + ### Added - **OMB M-25-22 AI acquisition-lifecycle tracking — `evidentia ai-gov diff --git a/README.md b/README.md index 03562b56..f251c8d5 100644 --- a/README.md +++ b/README.md @@ -23,7 +23,7 @@ Code of Conduct OpenSSF Best Practices OpenSSF Scorecard - CLI↔GUI parity + CLI↔GUI parity

@@ -88,8 +88,9 @@ See it first, no install — a self-hosted [asciinema](https://asciinema.org/) r ## Features -- **OSCAL-native end-to-end** — Ingest NIST OSCAL catalogs (Catalog 1.2.1); emit OSCAL Assessment Results + Plan-of-Action-and-Milestones (POA&M). Ready for FedRAMP 20x machine-readable submissions. -- **Cryptographic evidence chain** — Sigstore keyless signing on Assessment Results; PEP 740 attestations on every published wheel; SLSA Provenance v1 on the container; CycloneDX 1.6 SBOM on every GitHub Release. +- **OSCAL-native end-to-end** — Ingest NIST OSCAL catalogs (Catalog 1.2.1); emit OSCAL Assessment Results + Plan-of-Action-and-Milestones (POA&M) on the established FedRAMP Rev 5 package rail. +- **FedRAMP CR26 machine-readable SDR emission** — `evidentia conmon ksi` emits the CR26 Security Decision Record `keySecurityIndicators` block (10 families / 46 KSIs) conformant to FedRAMP's official 2026-06-24 schemas (vendored at pinned upstream SHAs, drift-watched weekly) — the first production-grade open-source emitter of the CR26 SDR format. +- **Cryptographic evidence chain** — Sigstore keyless signing on Assessment Results; PEP 740 attestations on every published wheel; SLSA Provenance v1 on the container; CycloneDX 1.7 SBOM on every GitHub Release. - **96 framework catalogs + 16 crosswalks** — NIST 800-53 Rev 5 (full 1,196 controls + Low/Moderate/High/Privacy baselines), CSF 2.0, FedRAMP (Rev 5 baselines + CR26 Key Security Indicators), CMMC 2.0 L1/L2, OpenSSF OSPS Baseline (Maturity 1/2/3 + first public OSCAL serialization), ISO 27001:2022, EU AI Act, DORA, NIS2, GDPR, all 15 US state privacy laws, full FFIEC IT Examination Handbook, OCC Bulletin 2026-13a / FRB SR 26-02. Plus 16 inter-framework crosswalks. - **14 evidence collectors** — AWS, GitHub (including v0.10.6 OSPS conformance helpers), Postgres, MySQL, Oracle, SQLite, MS-SQL, Snowflake, Databricks, Okta, Vanta, Drata, BitSight, SecurityScorecard. All OCSF-aligned with `compliance_status` field. - **OCSF-aligned findings** — OCSF Compliance Finding (class_uid 2003) via `--format ocsf`; OCSF Detection Finding (class_uid 2004) via `--format ocsf-detection`. SARIF 2.1.0 for CI gates via `--format sarif`. CycloneDX 1.6 VEX via `--format cyclonedx-vex`. @@ -106,7 +107,7 @@ See it first, no install — a self-hosted [asciinema](https://asciinema.org/) r | Evidence collectors | 14 | | MCP tools | 13 | | OSCAL serializations | 1 (OpenSSF OSPS Baseline; more on the v0.11+ roadmap) | -| Test suite | 4,000+ tests; mypy strict; ruff clean | +| Test suite | 4,900+ tests; mypy strict; ruff clean | ## Documentation diff --git a/docs/ROADMAP.md b/docs/ROADMAP.md index fd9c0cd6..f43c2b84 100644 --- a/docs/ROADMAP.md +++ b/docs/ROADMAP.md @@ -85,8 +85,9 @@ leads with the CLI. Evidentia hosts no state. Best fit for a CLI / library-shaped tool. - **Local writable GUI demo (pending a store-seeder):** a one-command `docker compose -f docker-compose.demo.yml up` with a seeded store + mock collectors. - The web console now surfaces ~98% of coverable CLI leaves; this is gated only on a - store-seeder. + The web console now surfaces 93.4% of coverable CLI leaves (the 7 api-only leaves are + the v0.11 federal AI-governance verbs plus two chrome-surfaced utilities); this is + gated only on a store-seeder. - **Not planned:** a public, hosted, *stateful* backend demo. A credentialed GRC backend exposed publicly is a real security surface (SSRF / secret exfiltration / prompt-injection via collected evidence — the class the `--block-private-ips` hardening addresses). A durable diff --git a/docs/api-stability.md b/docs/api-stability.md index a60be8c7..8215182e 100644 --- a/docs/api-stability.md +++ b/docs/api-stability.md @@ -456,7 +456,7 @@ The release pipeline validates API stability via: 1. **Type checking** (mypy strict): catches signature changes that would break callers -2. **Test suite** (2747+ tests): exercises public interfaces +2. **Test suite** (4,900+ tests): exercises public interfaces against expected behavior 3. **Import smoke test**: `scripts/check_imports.py` (reserved for v1.0) will validate that all documented entry points diff --git a/docs/cli-gui-parity.yaml b/docs/cli-gui-parity.yaml index 527bc14c..ef9d6d8e 100644 --- a/docs/cli-gui-parity.yaml +++ b/docs/cli-gui-parity.yaml @@ -36,6 +36,13 @@ # GUI coverage = full / (full + api-only + cli-only) = 13 / 98 = 13.3%. # (The "~30% GUI reaches the CLI" figure was API breadth: 55/98 leaves have an # API; 13 are now surfaced as a dedicated GUI screen.) +# +# Current distribution (2026-07-16, post-v0.11 Wave 2): 99 full / 7 api-only / +# 0 cli-only / 11 exempt (117 leaves) = 93.4% GUI coverage. The 7 api-only rows +# are the five v0.11 federal AI-governance verbs (`ai-gov set-practice` + +# `ai-gov acquisition *`) pending their console pass, plus the two +# chrome/index-surfaced utilities (`version`, `init`). Rendered view: +# docs/parity-coverage.md (generated by check_parity.py --emit-coverage-md). version: 1 # Debt-ratchet floor: the count of `cli-only` rows at seed time. check_parity.py diff --git a/docs/enterprise-grade.md b/docs/enterprise-grade.md index 8256988c..f7bd3efa 100644 --- a/docs/enterprise-grade.md +++ b/docs/enterprise-grade.md @@ -47,7 +47,8 @@ posture has advanced materially across the v0.8.x–v0.10.x line: cross-platform Hypothesis property tests (v0.10.12) harden the untrusted-input parsers and credit the OpenSSF Scorecard "Fuzzing" check (advances **M1**'s intent via coverage-guided fuzzing; classic - mutation testing remains unscheduled). + mutation testing runs weekly out-of-band via `mutmut.yml` without an + enforced kill-rate gate). - **WORM evidence store + retention / legal-hold** (v0.9.6) provide first-class anti-tamper evidence lineage at the product layer, advancing **H8** and **H13** beyond operator-only responsibility. @@ -60,11 +61,13 @@ posture has advanced materially across the v0.8.x–v0.10.x line: rescinded M-24-10), and a signed Control↔Threat traceability **OSCAL profile** (v0.10.11 — partially addresses **L4**'s "not OSCAL Profile format" note). -- **CLI↔GUI parity** — the local web console reaches ~98% parity across - 22 consoles (v0.10.12), with a documented web-console security model and +- **CLI↔GUI parity** — the local web console reaches 93.4% parity across + 22 consoles (99 of 106 coverable CLI leaves; the 7 api-only leaves are + the v0.11 federal AI-governance verbs plus two chrome-surfaced + utilities), with a documented web-console security model and a default-on SSRF / anti-DNS-rebind network guard across collectors, integrations, and the Tableau publish path. -- **95 bundled framework catalogs** across four redistribution tiers +- **96 bundled framework catalogs** across four redistribution tiers (updates **L4**'s "82 frameworks" count). - **Release + CI hardening (v0.10.13–v0.10.17)** — a distroless **Docker Hardened Images** runtime base (v0.10.16), **air-gapped DSSE / @@ -128,9 +131,9 @@ evidence lineage). | # | Item | v0.7.0 status | |---|---|---| -| M1 | Mutation testing ≥ 65% | ❌ Not currently run in CI; planned for v0.8 | +| M1 | Mutation testing ≥ 65% | ❌ Not run in CI at v0.7.0. *Later: weekly out-of-band mutmut CI runs since v0.8.2 (`mutmut.yml`: Sundays 03:00 UTC + dispatch); the ≥65% kill-rate gate is not yet enforced.* | | M2 | Air-gapped deployment mode | ✅ `--offline` flag refuses network egress; GPG signing works offline | -| M3 | Multi-cloud + SaaS support | ⚠️ AWS + GitHub in v0.7.0; Okta, Azure, GCP planned | +| M3 | Multi-cloud + SaaS support | ⚠️ AWS + GitHub at v0.7.0. *Later: grown to 14 collectors (adds Okta, 5 SQL adapters, Databricks, Snowflake, Vanta, Drata, BitSight, SecurityScorecard); Azure, GCP remain roadmap.* | | M4 | Performance benchmarks | ⚠️ Not documented; ad-hoc testing on 10k-resource sample | | M5 | Dry-run mode | ✅ `dry_run=True` kwarg on all collectors; `--dry-run` CLI flag | | M6 | Custom OSCAL finding fields | ✅ `back-matter.resources[].props[]` with Evidentia-namespaced prop extensions | diff --git a/docs/gemara-mapping.md b/docs/gemara-mapping.md index 51d624a3..3d6c2bd6 100644 --- a/docs/gemara-mapping.md +++ b/docs/gemara-mapping.md @@ -52,7 +52,7 @@ Adopters (as of 2026-05-23): | Gemara component | Evidentia surface | Notes | |---|---|---| -| **Catalog: Control** | `evidentia_core.models.catalog.ControlCatalog` (95 bundled — NIST 800-53, SOC 2, ISO 27001 stub, CIS, CMMC, PCI DSS, EU GDPR, EU AI Act, …) | Tier-classified (A/B/C/D — see `ATTRIBUTION.md`); JSON or YAML format (v0.10.3+ via `docs/contributing-a-catalog.md`). 1:1 conceptual match. | +| **Catalog: Control** | `evidentia_core.models.catalog.ControlCatalog` (96 bundled — NIST 800-53, SOC 2, ISO 27001 stub, CIS, CMMC, PCI DSS, EU GDPR, EU AI Act, …) | Tier-classified (A/B/C/D — see `ATTRIBUTION.md`); JSON or YAML format (v0.10.3+ via `docs/contributing-a-catalog.md`). 1:1 conceptual match. | | **Catalog: Capability** | partial — `Control.implementation_status` enum (`implemented` / `partial` / `planned` / `missing`) | Evidentia tracks implementation state per control, not as a separate capability registry. Could be promoted to a first-class Capability model if Gemara conformance becomes a goal. | | **Catalog: Principle** | no direct equivalent | Evidentia treats principles as "tier-D obligations" inside `ObligationCatalog` (statutes, regulatory edicts). Functionally adjacent but not 1:1. | | **Catalog: Risk** | `evidentia_ai.risk_statements.RiskStatement` (NIST SP 800-30 Rev 1) + `evidentia_core.models.governance.AISystemClassification` (EU AI Act tiers, NIST AI RMF) | Risk catalogs in Evidentia are generated, not pre-defined — the LLM-driven `RiskStatementGenerator` produces statements from gaps + system context. Pre-defined risk catalogs are a v0.11+ candidate. | diff --git a/docs/integration-survey.md b/docs/integration-survey.md index 10e4c5d5..83ae54d5 100644 --- a/docs/integration-survey.md +++ b/docs/integration-survey.md @@ -34,17 +34,17 @@ architectural patterns — observed in peer projects and emerging standards |---|---| | **OSCAL 1.2.1** | First-class emit: catalog, profile, assessment-results, POA&M, CONMON state. Import: catalog, profile, component-definition (ingest-only). SCR notification is RFC-0007 flat JSON, not an OSCAL model | | **Supply chain** | CycloneDX 1.7 SBOM; Sigstore/Rekor + PEP 740 attestations; cosign-signed container; SLSA L3 build provenance | -| **MCP server** | 8 tools (`list_frameworks`, `get_control`, `gap_analyze`, `gap_diff`, `conmon_list_cadences`, `conmon_next_due`, `conmon_check_state`, `conmon_health`); stdio / SSE / HTTP transports; per-tool CIMD scope enforcement; `SignedToolOutput` envelope | +| **MCP server** | 13 tools (`list_frameworks`, `get_control`, `gap_analyze`, `gap_diff`, `gap_analyze_sarif`, `collect_ocsf`, `tprm_vendor_list`, `poam_list`, `verify_signed_artifact`, `conmon_list_cadences`, `conmon_next_due`, `conmon_check_state`, `conmon_health`); stdio / SSE / HTTP transports; per-tool CIMD scope enforcement; `SignedToolOutput` envelope | | **REST API** | FastAPI; `/api/poam/*`, `/api/conmon/*`, `/api/gaps`, `/api/metrics`; auth middleware | | **CI** | Bundled composite GitHub Action (`gap-analysis`) | -| **Evidence collectors** | AWS (Config / Security Hub / IAM Access Analyzer), GitHub, Okta, ServiceNow, 5 SQL databases, Databricks, Snowflake | +| **Evidence collectors** | 14 total — AWS (Config / Security Hub / IAM Access Analyzer), GitHub, Okta, 5 SQL databases, Databricks, Snowflake, Vanta, Drata, BitSight, SecurityScorecard (ServiceNow's collect side rides the output integration) | | **Output integrations** | Jira, ServiceNow, Tableau, Power BI | **The structural observation.** Each collector today emits a framework-shaped finding. That couples collector count to framework count and is the root cause of the integration-breadth gap that [positioning-and-value.md](positioning-and-value.md) §6.3.2 honest-gap #3 -surfaces (8 evidence collectors vs. Vanta's 375+). §4.1 below is the +surfaces (14 evidence collectors vs. Vanta's 375+). §4.1 below is the structural fix — and every other integration in §3 gets cheaper once it is in place. @@ -185,7 +185,7 @@ That makes it **complementary to Evidentia, not competitive** — they are adjacent layers. The opportunity: - **Publish a thin Evidentia MCP connector/plugin into the - `grc-engineering-suite` marketplace.** Evidentia's MCP server (8 tools, + `grc-engineering-suite` marketplace.** Evidentia's MCP server (13 tools, signed output) is exactly the deterministic backend that toolkit's personas lack. This reaches a growing, well-aligned audience through a zero-setup distribution channel (`/plugin install`) that Evidentia @@ -287,56 +287,66 @@ v0.10.5 Phase 2 ships the conversion as an OSCAL catalog + profile pair for the baseline that defines OSS-project security hygiene — directly adjacent to where Evidentia sits in the §5.5 OSS GRC tier. -### 8.2 FedRAMP 20x readiness — KSI emission posture +### 8.2 FedRAMP 20x readiness — KSI emission posture (SHIPPED, v0.11) FedRAMP **20x** (the "machine-readable continuous-monitoring" pillar of -the FedRAMP modernization) entered its **Moderate pilot through -2026-03-31** and the **public rollout lands Q3 2026**. The central -artifact is **KSIs — Key Security Indicators** delivered per FedRAMP's -machine-readable schemas (**FRMR JSON** — FedRAMP Machine-Readable -Documentation; JSON-Schema definitions published in the new -`FedRAMP/schemas` repo, created 2026-06-09). OSCAL is not displaced — -it remains the Rev5 / RFC-0024 package format per NTC-0009; FRMR JSON -is the KSI delivery schema, not an OSCAL extension. Evidentia's -existing OSCAL emit + CONMON daemon (v0.9.6 first-mover) + SCR -notification surface (v0.9.7) compose into a near-direct KSI emitter, -with FRMR JSON as the emission target. v0.11 schedules the KSI binding -as a first-mover claim — the public rollout coincides with Evidentia's -federal-compliance theme window. Sources: fedramp.gov 20x pages; -github.com/FedRAMP/schemas; FedRAMP NTC-0009; RegScale OSCAL Hub. - -### 8.3 OpenVEX 0.2.0 emit (FIRST-MOVER among GRC engines) +the FedRAMP modernization) went live with the **CR26 consolidated +rules** (launched 2026-06-25 on the 2026-06-24 dated schema fileset; +mandatory for all 20x certifications 2027-01-01). The central artifact +is the **Security Decision Record (SDR)** — JSON per +`fedramp-security-decision-record-schema-2026-06-24.json` in +`FedRAMP/schemas`, whose `keySecurityIndicators` block carries the +per-KSI summaries the SDR-CSX-KSI rules make a MUST. The earlier +"FRMR JSON" framing is retired upstream. OSCAL is not displaced — it +remains the established Rev 5 package rail (NTC-0009's +machine-readable requirement is itself format-agnostic); the CR26 +SDR is a JSON schema, not an OSCAL extension. **Shipped in v0.11 Wave 2:** +`evidentia conmon ksi` emits the SDR `keySecurityIndicators` block +(10 families / 46 KSIs from the `FedRAMP/rules` consolidated dataset) +as a schema-valid full-SDR skeleton, with the official schemas +vendored at pinned upstream SHAs and a weekly drift sentinel +(`fedramp-schema-watch`). Sources: fedramp.gov 20x pages; +github.com/FedRAMP/schemas; github.com/FedRAMP/rules; FedRAMP +NTC-0009. + +### 8.3 OpenVEX 0.2.0 emit (planned — would be first among OSS GRC engines) **OpenVEX v0.2.0** (CNCF; JSON-LD; status enum `{not_affected, affected, fixed, under_investigation}`) is the emerging VEX format for vulnerability-exploitability statements. **Dependency- Track does NOT consume OpenVEX (only CycloneDX VEX)**; GUAC is the only -mature OSS consumer. Evidentia's planned **`evidentia vex emit`** verb -(v0.11) makes it the first OSS GRC engine to emit OpenVEX statements +mature OSS consumer. Evidentia's planned **`evidentia vex publish --rekor`** verb +(roadmap-tracked, not yet scheduled) would make it the first OSS GRC +engine to emit OpenVEX statements alongside its CycloneDX SBOM — the natural bridge between Evidentia's supply-chain artifacts (§2) and the wider supply-chain-VEX ecosystem. Coupled with a Grype subprocess wrapper (Grype emits neither `-o vex` nor `-o ocsf` natively — needs a thin wrapper) this closes the "vulnerability triage as compliance evidence" loop. -### 8.4 CISA Secure by Design (SbD) Pledge — first OSS signatory +### 8.4 CISA Secure by Design (SbD) Pledge — alignment statement The CISA **Secure by Design Pledge** (~100 signatories as of May 2026, **all commercial vendors**) has seven goals. **No OSS project has -signed.** Evidentia is uniquely positioned to be the first — every goal -maps onto existing Evidentia work (MFA-by-default, default deny on the -MCP surface, vulnerability disclosure, evidence of patching cadence, -etc.). v0.10.5 Phase 4 ships **SECURITY.md** + a **security.txt** + the -**SbD pledge SELFATTEST** (Markdown form per CISA template) — the first -OSS project to do so. Cross-stream signal: also closes 4 of the 8 -templates in skill v5.1's "compliance closure bundle". +signed.** Every goal maps onto existing Evidentia work +(MFA-by-default, default deny on the MCP surface, vulnerability +disclosure, evidence of patching cadence, etc.). v0.10.5 Phase 4 +shipped **SECURITY.md** + a **security.txt** + the **SbD pledge +alignment self-assessment** +([`SECURE-BY-DESIGN-PLEDGE.md`](SECURE-BY-DESIGN-PLEDGE.md) — a +voluntary alignment statement, explicitly **not a signatory claim**) — +the first OSS project to publish a per-goal SbD alignment +self-attestation. Formal Pledge signing is planned at the v1.0 +launch. Cross-stream signal: also closes 4 of the 8 templates in +skill v5.1's "compliance closure bundle". ### 8.5 SLSA Verification Summary Attestation (VSA) emit SLSA v1.0 introduces **VSAs** as the auditable third-party verification artifact (distinct from build provenance). The reference OSS emitter (`slsa-framework/slsa-verifier --emit-vsa`) exists but is rare in -practice. v0.11 schedules `evidentia vsa emit` so each release ships a +practice. The roadmap tracks `evidentia oscal vsa` (not yet scheduled; emit +targeted per SLSA v1.2) so each release would ship a VSA alongside its SLSA Provenance v1 + PEP 740 attestations — the verification-loop closure for high-trust operators. Cross-cite with NIST SSDF v1.2 IPD (Dec 17 2025 draft) PS.4 ("Robust and Reliable @@ -349,20 +359,26 @@ expectation. |---|---|---|---| | OSPS Baseline OSCAL conversion + upstream PR | First OSCAL conversion of the OpenSSF OSPS Baseline | v0.10.5 P2 | OSCAL catalog + profile in `marketplace/oscal/` | | OSPS-CONFORMANCE.md + CI gate | First OSS project to publish a self-conformance statement against OSPS Baseline | v0.10.5 P3 | `docs/OSPS-CONFORMANCE.md` + `.github/workflows/osps-conformance.yml` | -| CISA SbD Pledge SELFATTEST | First OSS signatory of the CISA Secure by Design Pledge | v0.10.5 P4 | `docs/SECURE-BY-DESIGN-PLEDGE.md` + signal upstream | -| OpenVEX emit | First OSS GRC engine to emit OpenVEX 0.2.0 | v0.11 P3 | `evidentia vex emit` CLI verb | -| FedRAMP 20x KSI emit | First OSS engine to emit Key Security Indicators per FedRAMP's machine-readable schemas (FRMR JSON). `kenithphilip/FedPy` (2026-05-28) is an evidence *collector* for KSIs (Ed25519-signed evidence, OSCAL output), not a KSI-submission *emitter* — the first-mover claim survives with that distinction stated | v0.11 P5 | `evidentia conmon ksi-emit` CLI verb | -| SLSA VSA emit | First OSS GRC engine to emit SLSA VSAs alongside provenance | v0.11 P4 | `evidentia vsa emit` CLI verb | +| CISA SbD Pledge alignment self-attestation | First OSS project to publish a per-goal SbD Pledge alignment self-attestation (a voluntary alignment statement, not a signatory claim; formal signing planned at v1.0) | v0.10.5 P4 | `docs/SECURE-BY-DESIGN-PLEDGE.md` | +| OpenVEX emit | Would be the first OSS GRC engine to emit OpenVEX 0.2.0 (planned; roadmap-tracked, not yet scheduled) | planned | `evidentia vex publish --rekor` CLI verb (planned) | +| FedRAMP CR26 SDR / KSI emit | **First production-grade OSS emitter** — and first from any established tool — of the CR26 Security Decision Record `keySecurityIndicators` block per the official 2026-06-24 schemas. Caveats stated: `Palladium-Innovations/fedramp-cr26-oscal` (CC0, experimental) emits draft CR26-from-OSCAL; `kenithphilip/FedPy` (2026-05-28) is an evidence *collector* for KSIs (Ed25519-signed evidence, OSCAL output), not an SDR *emitter*; RegScale (Excel exports + KSI→OSCAL pipeline) and Paramify (YAML + unofficial OSCAL) ship adjacent KSI tooling without official-schema conformance | v0.11 Wave 2 (SHIPPED) | `evidentia conmon ksi` CLI verb | +| OMB M-25-21 minimum-practice tracking | First structured tracking of the memo's seven §4(b) minimum practices with **CAIO waiver objects carrying the annual re-certification + 30-day OMB-report clocks** — structure the federal AI Use Case Inventory lacks entirely. (Scope note: the inventory itself carries 9 per-practice status fields to Evidentia's 7-practice model — the first-mover claim is the waiver-clock structure, never field count) | v0.11 Wave 2 (SHIPPED) | `evidentia ai-gov set-practice` CLI verb + API mirror | +| OMB M-25-22 acquisition lifecycle | First structured model of M-25-22's six §4 acquisition-lifecycle phases (Identification of Requirements → Contract Closeout) with the §4(a) high-impact determination tie-in — the federal inventory has no acquisition-phase structure at all | v0.11 Wave 2 (SHIPPED) | `evidentia ai-gov acquisition register\|list\|show\|set-phase` + API mirror | +| SLSA VSA emit | Would be the first OSS GRC engine to emit SLSA VSAs alongside provenance (planned; roadmap-tracked, not yet scheduled) | planned | `evidentia oscal vsa` CLI verb (planned) | ### 8.7 AWS OSCAL MCP cross-reference `awslabs/oscal-mcp-server` (Apache-2.0; `samples/` archived shape — not -a service) is the only other OSCAL-aware MCP server in the OSS -ecosystem as of May 2026. Its scope is **knowledge-tool only** — +a service) is one of several OSCAL-aware MCP servers in the OSS +ecosystem (alongside `oscal-compass/compliance-trestle-mcp` and CISO +Assistant's `cli/ca_mcp`). Its scope is **knowledge-tool only** — answers OSCAL schema + 800-53 / 800-171 catalog questions via embedded -references; does **not** emit OSCAL artifacts. Evidentia's MCP server -(§4.3) goes structurally further: emits component-definition / SSP / -assessment-results / POA&M / VEX / VSA. Worth cross-citing as the only +references; does **not** emit OSCAL artifacts. Evidentia goes +structurally further: its MCP server (§4.3) fronts a runtime engine +that emits OSCAL catalog / profile / assessment-results / POA&M +(plus SARIF / OCSF / CycloneDX-VEX gap outputs and the CR26 SDR KSI +block). Worth +cross-citing as an OSCAL-MCP precedent; not a competitor. See [positioning-and-value.md §5.5](positioning-and-value.md) for the same cross-reference in the OSS GRC tier table. @@ -370,13 +386,17 @@ GRC tier table. ### 8.8 ComplianceCow MCP cross-reference `compliancecow/cowmcp` (Apache-2.0 server; ComplianceCow backend -proprietary) is the **only other OSS GRC MCP server besides Evidentia's -`evidentia mcp serve`**. The shape difference: ComplianceCow MCP is a +proprietary) is **one of several OSS GRC MCP servers** — CISO +Assistant ships a full MCP server (`cli/ca_mcp`, expanded through +2026-06) and `oscal-compass/compliance-trestle-mcp` also exists, +alongside Evidentia's `evidentia mcp serve`. The shape difference: +ComplianceCow MCP is a thin facade over a SaaS backend (14 tools, all hitting the ComplianceCow API), whereas Evidentia is a self-contained runtime -engine. Worth cross-citing as the only other OSS GRC MCP — and as the -honest-landscape entry "two OSS GRC MCP servers exist as of May 2026, -both Apache-2.0". +engine. Worth cross-citing in the honest-landscape entry: several OSS +GRC MCP servers exist as of mid-2026 — Evidentia and ComplianceCow +(Apache-2.0), CISO Assistant's `ca_mcp` (AGPL), and the +authoring-focused trestle-mcp. ## 8.9 OSPS Baseline crosswalks (v0.10.6 Phase 5) diff --git a/docs/parity-coverage.md b/docs/parity-coverage.md index 5ef13b30..6cb4226d 100644 --- a/docs/parity-coverage.md +++ b/docs/parity-coverage.md @@ -2,17 +2,17 @@ > Generated by `scripts/check_parity.py --emit-coverage-md` from `docs/cli-gui-parity.yaml`. Do not edit by hand. -**GUI coverage: 98.0%** (99 of 101 coverable CLI leaves are surfaced as a dedicated GUI screen; `exempt` leaves are excluded). +**GUI coverage: 93.4%** (99 of 106 coverable CLI leaves are surfaced as a dedicated GUI screen; `exempt` leaves are excluded). ## Burndown | status | count | meaning | |---|---:|---| | full | 99 | CLI verb + API op + wired GUI route | -| api-only | 2 | API op exists; no dedicated GUI route yet | +| api-only | 7 | API op exists; no dedicated GUI route yet | | cli-only | 0 | no API + no GUI (ratchet floor: 0) | -| exempt | 10 | CLI-only by design (servers, local cache, offline verify, MCP) | -| **total** | **111** | every live CLI leaf | +| exempt | 11 | CLI-only by design (servers, local cache, offline verify, MCP) | +| **total** | **117** | every live CLI leaf | ## Per-command status @@ -82,6 +82,11 @@ | `ai-gov categorize-fips` | POST /api/ai-gov/systems/{system_id}/categorize-fips | /ai-gov | full | | `ai-gov set-omb-impact` | POST /api/ai-gov/systems/{system_id}/set-omb-impact | /ai-gov | full | | `ai-gov set-high-impact` | POST /api/ai-gov/systems/{system_id}/set-high-impact | /ai-gov | full | +| `ai-gov set-practice` | POST /api/ai-gov/systems/{system_id}/set-practice | — | api-only | +| `ai-gov acquisition register` | POST /api/ai-gov/acquisitions | — | api-only | +| `ai-gov acquisition list` | GET /api/ai-gov/acquisitions | — | api-only | +| `ai-gov acquisition show` | GET /api/ai-gov/acquisitions/{acquisition_id} | — | api-only | +| `ai-gov acquisition set-phase` | POST /api/ai-gov/acquisitions/{acquisition_id}/set-phase | — | api-only | | `catalog crosswalk` | GET /api/catalog/crosswalk | /catalog | full | | `catalog import` | POST /api/catalog/import | /catalog | full | | `catalog where` | GET /api/catalog/where | /catalog | full | @@ -119,6 +124,7 @@ | `evidence show` | GET /api/evidence/{lineage_id}/versions/{version} | /evidence | full | | `serve` | — | — | exempt | | `conmon watch` | — | — | exempt | +| `conmon ksi` | — | — | exempt | | `oscal verify` | POST /api/oscal/verify | /oscal | full | | `traceability emit` | POST /api/traceability/emit | /traceability | full | | `explain cache clear` | — | — | exempt | diff --git a/docs/positioning-and-value.md b/docs/positioning-and-value.md index 6890e8a9..258e8ac4 100644 --- a/docs/positioning-and-value.md +++ b/docs/positioning-and-value.md @@ -37,13 +37,14 @@ ## 1. Executive summary -> **Section summary.** Evidentia (v0.7.8, May 2026) is the only +> **Section summary.** Evidentia is the only > open-source, Python-first, library-first, OSCAL-native GRC tool -> with bundled multi-framework catalogs (88), supply-chain-hardened +> with bundled multi-framework catalogs (96), supply-chain-hardened > evidence integrity (Sigstore + SLSA L3 + PEP 740 + cosign-signed > container), and an air-gap-capable deployment model. It sits in a -> near-empty competitive niche, on the right side of **four** -> regulatory forcing functions converging in Q4 2026, and is +> near-empty competitive niche, on the right side of a stacked +> multi-year regulatory mandate runway (CMMC Phase 2 + EU CRA in +> Q4 2026; FedRAMP CR26 2027-01-01; NTC-0009 2027-11-01), and is > intellectually descended from the Kellogg et al. ASE 2020 > "Continuous Compliance" thesis (now extended by the Khatchadourian > March 2026 DFAH paper that names Evidentia's planned v0.8.0 flagship @@ -62,7 +63,7 @@ - **PCAOB's July 2024 Generative AI Spotlight (non-binding staff observations) + the 2024 AS 1105 amendments (technology-assisted analysis of electronic information) frame five AI-in-audit concerns; Evidentia maps its own capabilities onto all 5.** (1) Prompt + response capture — Evidentia v0.7.1 ✓; (2) Model-version tracking — v0.7.1 ✓; (3) Human-in-the-loop verification — DFAH addresses; (4) Immutable audit trails — DFAH addresses; (5) Explainability documentation — PRT addresses. **This mapping — Evidentia's own, not a PCAOB requirement — is among the strongest positioning claims Evidentia can make** in the v0.8.0 cycle (see §11). - **SR 11-7 superseded by SR 26-02 + OCC Bulletin 2026-13 (the "13a" PDF is the bulletin's guidance attachment) (April 17, 2026)** — and the new guidance **explicitly excludes generative AI and agentic AI**: *"Generative AI and agentic AI models are novel and rapidly evolving. As such, they are not within the scope of this guidance."* Banks are stranded with no regulatory framework for LLM deployments. **This is the precise opening for Evidentia's planned v0.7.9 model-risk module** to ship the SR-11-7-replacement-framework primitive ahead of any commercial vendor (see §7, §11). - **Positioning frame (revised 2026-05-25 per triple-validated procurement pressure-test): direct enumerated claims, no metaphor.** Three independent procurement-style reviewers (GPT-5.5 Fortune 500 procurement = 5/10; Grok-4.20-Multi-Agent federal-SI hostile TPRM = 2.5/10; Grok 4.3 heterodox marketing strategist = reject metaphor entirely) converged on the same diagnosis: the "Terraform/dbt of GRC" framing overclaims a multi-vendor consortium status a solo-vendor implementation cannot credibly own, and the "compliance kernel" alternative reads as developer-jargon to compliance buyers. The replacement positioning leads with six attestable supply-chain credentials (PEP 740 + Sigstore + SLSA L3 + cosign + CycloneDX 1.7 + OpenSSF Scorecard 6.5 + OpenSSF Silver MET + OSPS Baseline Maturity 2) + a concrete output enumeration (OSCAL + SARIF + OCSF Compliance Finding + OCSF Detection Finding + CycloneDX VEX) + a SOC 2 Type I in-progress signal + an honest acknowledgment of the OpenSSF Gold structural ceiling tied to the SOC 2 Type I program's segregation-of-duties controls. See §10 for the supporting messaging hierarchy. -- **Honest gaps** (intentionally surfaced, not buried): no customer-facing trust center, no questionnaire-fill AI, **6 evidence-collection surfaces** vs Vanta's 375+ integrations (Evidentia ships AWS, GitHub, Okta, ServiceNow, 5 SQL DB adapters, Databricks, Snowflake; Azure + GCP extras declared in pyproject.toml but currently lack implementations — see §5/§6 honest gap), no auditor partnerships, no Forrester Wave / Gartner inclusion, no TPRM module (planned v0.7.9), no risk register / ERM primitive, no published reference customers, no formal SOC 2 of Evidentia itself, no IPO/exit narrative for vendor-longevity-paranoid buyers. Each has a planned remediation or an explicit "won't fix" rationale (see §6). +- **Honest gaps** (intentionally surfaced, not buried): no customer-facing trust center, no questionnaire-fill AI, **14 evidence collectors** vs Vanta's 375+ integrations (AWS, GitHub, Okta, 5 SQL DB adapters, Databricks, Snowflake, Vanta, Drata, BitSight, SecurityScorecard; Azure + GCP remain roadmap — see §5/§6 honest gap), no auditor partnerships, no Forrester Wave / Gartner inclusion, no published reference customers, no formal SOC 2 of Evidentia itself, no IPO/exit narrative for vendor-longevity-paranoid buyers. Each has a planned remediation or an explicit "won't fix" rationale (see §6). - **Intellectual home**: an updated 30+ person community across NIST/GSA, OpenSSF, CISA-SBOM, Stanford CodeX, Bath/PRT, OSCAL Compass + OSCAL Foundation + Resilient Cyber / fwd:cloudsec / GRC-Engineering crowd. **NIST CSWP 53 *Charting the Course for NIST OSCAL*** (Iorga et al., December 2025 IPD) is now the canonical OSCAL strategic-direction citation, replacing older OSCAL references. Two outreach actions reach the entire community in one quarter (see §12). - **The 12-month direction** that compounds Evidentia's advantages: ship a DFAH-style determinism harness for risk-statement generation (academic prior art exists — Khatchadourian arXiv 2601.15322 March 2026 — but **no commercial vendor offers this**); publish the bundled multi-framework crosswalk dataset as a standalone artifact (`evidentia-catalogs`); add canonical scanner-JSON-to-OSCAL mappers; ship an MCP server (note: first-mover window for OSCAL primitives is **closed** as of Q1 2026 — IBM trestle-MCP, AWS Labs OSCAL MCP, Vanta MCP, Drata MCP, Optro MCP all live; **wide open** for SR-11-7 / FFIEC / SBOM / continuous-monitoring MCPs); ship the v0.7.9 federal-compliance overlay (TPRM + model risk + 7 new catalogs + governance primitives); pursue OSCAL Foundation membership; cut v0.8.x around AI evidence validation + multimodal evidence support; publish the **first standardized GRC LLM eval suite to Hugging Face** (HF Hub keyword searches confirm no substantive/eval-grade GRC datasets — zero OSCAL or SOC 2 hits; the lone NIST 800-53 hit is a sub-1K control-text dump — genuine first-in-class opportunity; see §13). @@ -127,11 +128,13 @@ Meanwhile, the compliance workload keeps growing. A single fintech or healthcare --- -## 3. Current capabilities (v0.10.12) +## 3. Current capabilities -> **Section summary.** As of v0.10.12 (June 2026), Evidentia ships **22 Typer +> **Section summary.** As of the v0.11.0 release prep (July 2026), Evidentia ships **22 Typer > CLI commands** (18 groups + 4 utility verbs), **23 REST router modules**, a -> full multi-console React web app (**22 consoles at ~98% CLI parity**) that +> full multi-console React web app (**22 consoles at 93.4% CLI parity** — 99 of +> 106 coverable CLI leaves; the 7 api-only leaves are the v0.11 federal +> AI-governance verbs plus two chrome-surfaced utilities) that > renders its own version live from the API, **13 read-mostly MCP tools**, six > Python packages with public APIs + the evidentia-ui frontend that bundles into > the API server, **14 credentialed evidence collectors** @@ -141,7 +144,7 @@ Meanwhile, the compliance workload keeps growing. A single fintech or healthcare > **4 output integrations** (Jira, ServiceNow, Tableau publish, Power BI > publish), AI-powered risk-statement + control-explanation generation > with full GenerationContext provenance (model + temperature + -> prompt-hash + run-id), **95 bundled framework catalogs** (4 redistribution +> prompt-hash + run-id), **96 bundled framework catalogs** (4 redistribution > tiers), bundled crosswalks, and **eight gap-output formats** including OSCAL > 1.2.1 Assessment Results with embedded SHA-256 evidence digests + Sigstore/Rekor > or GPG signing. Every release carries six independently attestable supply-chain @@ -165,7 +168,7 @@ for the full subcommand set. | `evidentia doctor [--check-air-gap]` | System diagnostics — LLM connectivity, web-UI availability, and the per-subsystem air-gap posture report. | | `evidentia serve` | Launch the FastAPI server + bundled React SPA (defaults to `127.0.0.1:8000`). | | `evidentia gap {analyze,diff}` | Run gap analysis (eight output formats) and diff two snapshots (opened / closed / severity-changed / unchanged). | -| `evidentia catalog {list,show,crosswalk,import,license-info}` | Browse the 95 bundled framework catalogs, show a control, map controls across frameworks, import licensed Tier-C content, inspect license posture. | +| `evidentia catalog {list,show,crosswalk,import,license-info}` | Browse the 96 bundled framework catalogs, show a control, map controls across frameworks, import licensed Tier-C content, inspect license posture. | | `evidentia risk generate` | LLM-generated NIST SP 800-30 risk statements with structured output. | | `evidentia explain control ` | Plain-English, LLM-generated translation of any control's text (on-disk cached). | | `evidentia integrations {jira,servicenow,tableau,powerbi}` | Output integrations — push gaps to Jira / ServiceNow and publish to Tableau / Power BI. | @@ -196,9 +199,12 @@ The web UI is a thin client over the REST API: every operator surface in the browser is backed one-to-one by an API endpoint, and the React app renders its own version live from the API rather than hardcoding it. The 23 router modules are the full operator surface — there is no -admin-only or paid-tier API. CLI ↔ GUI parity is roughly **98% across -~22 in-browser consoles**: every CLI command group except the -server-side server/`mcp`/`eval` plumbing has an equivalent console, and +admin-only or paid-tier API. CLI ↔ GUI parity is **93.4% across the 22 +in-browser consoles** (99 of 106 coverable CLI leaves; the 7 api-only +leaves — the v0.11 federal AI-governance verbs plus the two +chrome-surfaced utilities — have API endpoints but no dedicated console +yet): every CLI command group except the server-side +server/`mcp`/`eval` plumbing has an equivalent console, and the operator-facing surface is fully open source. The 23 routers cover: @@ -233,8 +239,9 @@ live in the 25-page [Guides](wiki/2-guides/index.md) section. `evidentia serve` launches a sidebar-driven, multi-console single-page app — **22 non-demo consoles** (Home plus 21 grouped consoles), each backed one-to-one by a REST endpoint and calling the same engine as the CLI. There is no UI-only -analysis path and no admin-only or paid-tier surface. CLI↔GUI parity is roughly -**98% across ~22 consoles**: every CLI command group except the server-side +analysis path and no admin-only or paid-tier surface. CLI↔GUI parity is +**93.4% across the 22 consoles** (7 api-only leaves pending their console +pass): every CLI command group except the server-side plumbing (`serve` / `mcp` / `eval`) has an equivalent console. The app renders its own version live from the API (`/api/health`) rather than hardcoding it. @@ -351,22 +358,22 @@ network guard described in [§3.7](#37-supply-chain--security-hardening). - **Tableau publish** (`[tableau]` extra) — publishes 3 datasets to Tableau Server / Cloud as CSV-based data sources: `evidentia-gaps` (one row per ControlGap), `evidentia-risks` (NIST SP 800-30 shape with AI-provenance fields), and `evidentia-collection-runs` (the CollectionContext audit trail). PAT auth via the `TABLEAU_PAT_NAME` + `TABLEAU_PAT_SECRET` env vars (never accepted as flag values). `tableauserverclient>=0.30` (pure Python). A `.hyper`-extract publish path remains a roadmap enhancement under a separate `[tableau-hyper]` extra (it would require the heavyweight `tableauhyperapi` native binary). - **Power BI publish** (`[powerbi]` extra) — pushes the same 3 datasets as Power BI Push Datasets via the REST API + Azure AD service-principal OAuth2 (MSAL Python). Full-refresh semantics by default (clear-then-push), 10,000-row batching per Power BI's documented limit, and schema-declared dataset auto-creation via `ensure_dataset` (idempotent re-runs). Auth: a service principal with `Dataset.ReadWrite.All`; the client secret is read from `POWERBI_CLIENT_SECRET` server-side and never appears in request bodies. `msal>=1.31` driver. -### 3.6 Bundled framework catalogs (95 total, four redistribution tiers) +### 3.6 Bundled framework catalogs (96 total, four redistribution tiers) -**95 catalogs** verified by the canonical +**96 catalogs** verified by the canonical `scripts/check_doc_counts.py` gate, which counts every `frameworks:` entry in the catalogs manifest (`packages/evidentia-core/src/evidentia_core/catalogs/data/frameworks.yaml`) — -exactly the list `FrameworkRegistry.list_frameworks()` returns. Of the 95, -**73 ship as real catalogs** (control/technique/vulnerability/obligation +exactly the list `FrameworkRegistry.list_frameworks()` returns. Of the 96, +**74 ship as real catalogs** (control/technique/vulnerability/obligation text bundled) and **22 are licensed placeholder stubs** (public clause numbering + neutral titles only; you load your licensed copy via `evidentia catalog import`). The 22 placeholders are exactly the Tier-C licensed frameworks; Tiers A, B, and D contain no placeholders. The four redistribution tiers (see [contributing a catalog](contributing-a-catalog.md) -for the licensing model) sum to the stated total — **49 + 4 + 22 + 20 = 95**: +for the licensing model) sum to the stated total — **50 + 4 + 22 + 20 = 96**: -- **Tier A — Verbatim public-domain / open-licensed (49 frameworks)**: +- **Tier A — Verbatim public-domain / open-licensed (50 frameworks)**: **39 US federal** — NIST SP 800-53 Rev 5 baselines (Low / Moderate / High / Privacy, 5.2.0), 800-171 r2/r3, 800-172, CSF 2.0, AI RMF 1.0, Privacy Framework 1.0, SSDF 800-218; FedRAMP Rev 5 baselines (Low / @@ -629,7 +636,7 @@ The OSS GRC tier kept surging through 2025-26. **None of these is a permissively | **trycompai/comp** (Comp AI) | **1,620 stars (2026-06-10); daily commits**; **$2.6M pre-seed** announced 2025-08-01 (co-led by OSS Capital + Grand Ventures); 700+ companies (per trycomp.ai, 2026-06-10) | Vanta/Drata alternative; AI-native positioning; TypeScript/Next.js | AGPL-3.0 + commercial | Targets SaaS-startup SOC 2 self-serve market (different ICP); top-of-funnel awareness threat | | **GRCEngClub/claude-grc-engineering** | 295 stars (2026-06-10); created Dec 2025; latest tag v0.0.4 (Apr 2026) | Claude Code plugin marketplace — 63 plugins / 98 skills / 257 commands (persona + framework + connector plugins); a prompt-orchestration layer, **not** a runtime engine | MIT core (bundles CC BY-ND 4.0 SCF crosswalk data + CC BY-SA 4.0 CIS content) | **Complementary, not competitive** — a Claude-Code orchestration layer that explicitly is not an engine; its `grc-engineer` hub shells out to external tools. Interop opportunity: publish an Evidentia MCP plugin into its `grc-engineering-suite` marketplace. See [integration-survey.md](integration-survey.md) §5 | | **compliancecow/cowmcp** (ComplianceCow MCP) | Verified 2026-06-10; small & new project | MCP server fronting ComplianceCow's commercial GRC SaaS — exposes ~14 tools (control inventory, evidence retrieval, finding query, framework crosswalk lookup) for Claude / ChatGPT consumption | Apache-2.0 (server); ComplianceCow backend proprietary | **One of several OSS GRC MCP servers** — CISO Assistant ships a full MCP server (`cli/ca_mcp`; POC PR #1831 merged 2025-04-13, expanded through 2026-06-03) and oscal-compass/compliance-trestle-mcp also exists, alongside Evidentia's `evidentia mcp serve`. Different shape — MCP is a thin facade over a SaaS backend, whereas Evidentia is a runtime engine. Interop angle: cross-cite in the honest-landscape entry | -| **awslabs/oscal-mcp-server** (AWS OSCAL MCP) | Verified 2026-05-24 via awslabs repo; archived `samples/` shape — not a service | MCP knowledge tool — answers OSCAL schema + 800-53 / 800-171 catalog questions via embedded references | Apache-2.0 | **Knowledge tool only — not a runtime emitter.** Evidentia goes further: emits OSCAL component-definition / SSP / assessment-results / POA&M / VEX / VSA. Useful upstream reference; not a competitor | +| **awslabs/oscal-mcp-server** (AWS OSCAL MCP) | Verified 2026-05-24 via awslabs repo; archived `samples/` shape — not a service | MCP knowledge tool — answers OSCAL schema + 800-53 / 800-171 catalog questions via embedded references | Apache-2.0 | **Knowledge tool only — not a runtime emitter.** Evidentia goes further: a runtime engine that emits OSCAL catalog / profile / assessment-results / POA&M (plus SARIF / OCSF / CycloneDX-VEX gap outputs and the CR26 SDR KSI block). Useful upstream reference; not a competitor | | **simplerisk/simplerisk** | 101 stars; latest 2026-04-25 | PHP-based risk register; ~12 years old; commercial hosted version exists | Mozilla Public License 2.0 | Narrow shape — risk register only | | **Probo** | YC X25 batch (Spring 2025) | OSS SOC 2/ISO 27001/HIPAA platform | MIT | Service-augmented; small team | | **Openlane** (`theopenlane.io`) | Apache-2.0 | OSS GRC platform | Apache-2.0 | Different shape (platform vs library) | @@ -644,7 +651,7 @@ The OSS GRC tier kept surging through 2025-26. **None of these is a permissively | **open-policy-agent/opa** | 11,674 stars; CNCF Graduated | Rego-based; universal policy-as-code substrate | Apache-2.0 | **Could integrate** | | **cedar-policy/cedar** | 1,447 stars; AWS-led; Rust-based | Authorization policy language with Lean-verified symbolic compiler | Apache-2.0 | **Could integrate** | | **guacsec/guac** | 1,485 stars; OpenSSF | Aggregates SBOM/provenance into a graph | Apache-2.0 | Complementary supply-chain GRC angle | -| **GovReady-Q** | 210 stars | Django-based self-service GRC questionnaire app | GPL-3.0 | Effectively defunct (no meaningful work since 2024); Evidentia walks into the vacuum | +| **GovReady-Q** | 210 stars | Django-based self-service GRC questionnaire app | GPL-3.0 | Dormant, not defunct (default-branch commits end 2023-06-11; last push 2024-12-10); Evidentia walks into the vacated niche | | **opencontrol/compliance-masonry** | 377 / maintenance | Pre-OSCAL YAML "OpenControl" docs builder | Apache-2.0 | Predecessor of OSCAL; stale | | **RS-Credentive/oscal-pydantic** | 23 / 5; last push 2024-04-06 (2+ years stale) | Pydantic models autogenerated from OSCAL schemas | Apache-2.0 | Functionally archived; Evidentia could supersede | | **Eramba** ([eramba/docker](https://github.com/eramba/docker)) | 70 stars (docker repo); latest 2026-04-22 | **Still offers Community Edition (free forever, on-premise) + paid Enterprise tier** (verified 2026-05-18 via direct WebFetch of eramba.org). Community + Enterprise both support unlimited users + unlimited frameworks. The v0.7.8 baseline claim that Eramba had "shifted to closed-source application Q1 2026" was **inaccurate** and is corrected here. | Open-source Community + commercial Enterprise dual-license | Active competitor; PHP/Laravel stack; oriented to ISO 27001 + SOC 2 audit workflows; weaker on OSCAL + AI-quality than Evidentia | @@ -726,7 +733,7 @@ Evidentia's existing MCP / CIMD scope-enforcement substrate. > **eight genuine differentiators** versus every commercial tier — > the original seven (open-source + Python + library-first; > OSCAL-native; AI-optional; air-gap-capable; supply-chain-hardened -> on the GRC tool itself; composite GitHub Action / CI-native; 88 +> on the GRC tool itself; composite GitHub Action / CI-native; 96 > framework catalogs bundled) plus **a 5/5 capability mapping onto > PCAOB AS 1105 + GenAI Spotlight concerns** > via v0.7.1 audit trails + planned v0.8.0 DFAH + PRT (Evidentia's own @@ -735,14 +742,13 @@ Evidentia's existing MCP / CIMD scope-enforcement substrate. > mapping, MCP server (commercial vendors caught up Q1 2026), evidence > collection from cloud APIs, CycloneDX SBOM, Sigstore signing, and > conventional commit + semver discipline. It is **honestly behind** -> on the same ten surfaces as v0.7.0 (trust center; AI questionnaire -> fill; integration breadth — 8 evidence-collection surfaces vs -> Vanta's 375+; auditor partnerships; analyst inclusion; TPRM -> module — planned v0.7.9; ERM/risk-register primitive; reference -> customers; formal SOC 2 attestation; IPO/exit longevity narrative) -> plus a NEW v0.7.8-surfaced gap: **two `[azure]` and `[gcp]` -> optional extras declared in pyproject.toml without backing -> implementations** (queued for v0.7.9 batch fix). +> on the surviving honest gaps (trust center; AI questionnaire +> fill; integration breadth — 14 evidence collectors vs +> Vanta's 375+; auditor partnerships; analyst inclusion; reference +> customers; formal SOC 2 attestation; IPO/exit longevity +> narrative — TPRM and the risk-register primitive have since +> shipped, and the once-flagged `[azure]`/`[gcp]` phantom extras +> were removed). ### 6.1 The seven genuine differentiators (vs. all four commercial tiers) @@ -754,7 +760,7 @@ Evidentia's existing MCP / CIMD scope-enforcement substrate. | 4 | **Air-gap-capable** | None deploy on-prem or air-gapped | OneTrust on-prem deprecated; rest cloud-only | All cloud-only | All cloud-telemetry-dependent by definition | | 5 | **Sigstore + PEP 740 + SBOM applied to the GRC tool itself** | None publish signed SBOM for their platform | None | None | Parity-ish with Chainguard's signed-image philosophy, but Evidentia applies it to *the GRC tool*, not just images. **Genuine differentiator at the meta-layer.** | | 6 | **Composite GitHub Action / CI-native** | SaaS dashboards; not pipeline-native | Workflow apps; not CI primitives | Same | Snyk/Aqua/Sysdig CI-native for *security scanning*; Evidentia CI-native for *compliance evidence* | -| 7 | **88 framework catalogs bundled** (95+ planned for v0.7.9 with FFIEC + OCC + SR 11-7 additions) | Vanta 35+ ([Vanta frameworks](https://www.vanta.com/products/additional-frameworks)), Drata 20+, Secureframe ~12, Sprinto 200+ (claim — but with documented uneven depth — see §6.4), Comp AI 25+ | OneTrust + ServiceNow + IBM cover hundreds via paid licensed regulatory libraries | N/A (trust center vendors don't ship catalogs) | RegScale claims 60+ (parity); CISO Assistant 130+ claimed (parity) | +| 7 | **96 framework catalogs bundled + 16 crosswalks** (machine-gated count via `scripts/check_doc_counts.py`) | Vanta 35+ ([Vanta frameworks](https://www.vanta.com/products/additional-frameworks)), Drata 20+, Secureframe ~12, Sprinto 200+ (claim — but with documented uneven depth — see §6.4), Comp AI 25+ | OneTrust + ServiceNow + IBM cover hundreds via paid licensed regulatory libraries | N/A (trust center vendors don't ship catalogs) | RegScale claims 60+ (parity); CISO Assistant 130+ claimed (parity) | | 8 | **5/5 capability mapping onto PCAOB AS 1105 + GenAI Spotlight concerns** (Evidentia's own mapping onto the Spotlight's non-binding staff observations; NEW v0.7.8 differentiator) | None publish an equivalent mapping | None | None | None | ### 6.1.A The moat trinity (v0.9.6 sharpened framing — May 2026) @@ -763,13 +769,20 @@ The v0.9.5 Q3 2026 quarterly resync sharpened the differentiation story by collapsing the seven-vs-eight differentiator list into three concrete moats that compound rather than enumerate. **The moat trinity**: -1. **OSCAL emission** — Evidentia is the only OSS GRC tool that - emits catalog / profile / assessment-results / plan-of-action- - and-milestones (v0.9.6 ships OSCAL 1.2.1 via the - `OSCAL_SCHEMA_VERSION` constant; trestle-compatible) **as - first-class output**, not via an exporter add-on. The Sept 30 - 2026 FedRAMP machine-readable mandate makes this the load- - bearing differentiator for federal-adjacent operators. +1. **Machine-readable federal emission (OSCAL + CR26 JSON)** — + Evidentia is the only OSS GRC tool that emits catalog / profile / + assessment-results / plan-of-action-and-milestones (OSCAL 1.2.1 + via the `OSCAL_SCHEMA_VERSION` constant; trestle-compatible) **as + first-class output**, not via an exporter add-on — and, as of + v0.11, the first production-grade OSS emitter of the FedRAMP CR26 + Security Decision Record `keySecurityIndicators` block (official + 2026-06-24 schemas). Two federal machine-readable rails make this + the load-bearing differentiator: FedRAMP CR26's JSON-MUST rules + (FRC-CSO-JSN; CR26 mandatory for all 20x certifications + 2027-01-01) and NTC-0009's comprehensive machine-readable + requirement for Rev 5 Class D (High) by 2027-11-01 + (format-agnostic — it names no required format; OSCAL is the + established package rail). 2. **DFAH / PRT determinism** — auditor-defensible reproducibility on every AI-generated artifact. Khatchadourian arXiv 2601.15322 (v2 March 2026) names this by exact terminology; @@ -816,15 +829,15 @@ have." ### 6.2 Where Evidentia is at parity (don't oversell) - **AI-assisted control mapping** — Vanta AI Agent, SafeBase AI, RegScale all do this. Evidentia at parity, not advantage. -- **Multi-framework crosswalking** — Hyperproof's Hyperintelligence and Drata's framework engine are mature. Evidentia at 95 catalogs is at parity in coverage; quality of the crosswalk graph needs more validation. -- **Evidence collection from cloud APIs** — Tier 1 vendors and CSPM tools both do this well. Evidentia's collectors (AWS, GitHub Dependabot, IAM Access Analyzer) are at parity for the integrations shipped, behind on coverage breadth. +- **Multi-framework crosswalking** — Hyperproof's Hyperintelligence and Drata's framework engine are mature. Evidentia at 96 catalogs is at parity in coverage; quality of the crosswalk graph needs more validation. +- **Evidence collection from cloud APIs** — Tier 1 vendors and CSPM tools both do this well. Evidentia's 14 collectors (AWS, GitHub, Okta, 5 SQL adapters, Databricks, Snowflake, Vanta, Drata, BitSight, SecurityScorecard) are at parity for the integrations shipped, behind on coverage breadth. - **CycloneDX SBOM output** — Industry standard now (Anchore, Chainguard, Snyk). Parity, not ahead. - **Sigstore/Rekor signing of releases** — Increasingly table-stakes (Chainguard pioneered, Anchore + others adopting). Parity. - **Conventional commits + semver release automation** — Table-stakes in modern OSS. Parity. ### 6.3.1 Framework-count caveat (added v0.7.8) -Per Stream 1's findings on commercial GRC over-promising patterns: bare framework counts (e.g., "Sprinto 200+") hide deep unevenness — vendors typically have world-class SOC 2 / ISO 27001 coverage but checklist-only mapping with little automation for FFIEC, SR 11-7, FedRAMP, CMMC. PolicyCortex's CMMC analysis is explicit: SOC-2-grade evidence "is often not sufficient for a CMMC C3PAO assessor." Evidentia's 88 catalogs claim should always be qualified with **redistribution-tier accounting** (Tier A 31 frameworks verbatim public domain + Tier D 21 statutory + Tier C 20 licensed-stubs + Tier B 4 threat catalogs at the v0.7.0 base, +6 since) and the **catalog-import flow for Tier-C licensed copies** (`evidentia catalog import` for ISO 27001, PCI DSS 4.0.1, etc.). +Per Stream 1's findings on commercial GRC over-promising patterns: bare framework counts (e.g., "Sprinto 200+") hide deep unevenness — vendors typically have world-class SOC 2 / ISO 27001 coverage but checklist-only mapping with little automation for FFIEC, SR 11-7, FedRAMP, CMMC. PolicyCortex's CMMC analysis is explicit: SOC-2-grade evidence "is often not sufficient for a CMMC C3PAO assessor." Evidentia's 96 catalogs claim should always be qualified with **redistribution-tier accounting** (the four redistribution tiers per §3.6) and the **catalog-import flow for Tier-C licensed copies** (`evidentia catalog import` for ISO 27001, PCI DSS 4.0.1, etc.). ### 6.3.2 The ten honest gaps (intentionally surfaced) @@ -834,14 +847,14 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a |---|---|---| | 1 | **No customer-facing trust-center module** (SafeBase / Conveyor / Vanta Trust Center are now table-stakes for B2B SaaS buyers) | v0.8.x candidate — could be a `evidentia trust-center` static-site generator that emits Sigstore-attested trust pages from Evidentia's own evidence | | 2 | **No questionnaire-fill AI** (Conveyor's Sue, Drata/SafeBase, Vanta Questionnaire Automation are mature) | Won't ship — explicit positioning choice; Evidentia is library-first, not RFP-fill-first | -| 3 | **Integration breadth** (Vanta has 375+; Evidentia has 5 collectors today) | Roadmap: Okta, Azure, GCP collectors planned; community-pluggable collector pattern is documented | +| 3 | **Integration breadth** (Vanta has 375+; Evidentia has 14 collectors today) | Roadmap: Azure, GCP collectors planned; community-pluggable collector pattern is documented | | 4 | **No formalized auditor partnerships / no in-platform auditor handoff** | v0.8.x candidate — `evidentia oscal export --bundle-for-auditor` workflow + curated partnership list | | 5 | **No published Forrester Wave / Gartner Magic Quadrant inclusion** | Realistic only after a commercial sponsor exists; OSS projects don't typically get on these without one | -| 6 | **No third-party risk (TPRM) module** | Composable — Evidentia could ingest BitSight / SecurityScorecard data via collector pattern; not a v0.8 priority | -| 7 | **No risk register / ERM primitive** (AuditBoard, Riskonnect, MetricStream lead with ERM) | Honest framing: Evidentia's "risk statements" are NIST SP 800-30 narratives, not a Riskonnect-class ERM register. Could add a lightweight `evidentia risk register` in v0.9.x | +| 6 | ~~No third-party risk (TPRM) module~~ — **shipped v0.7.9** (`evidentia tprm`: vendor inventory, concentration risk, CAIQ/SIG questionnaires + BitSight / SecurityScorecard collectors) | Residual gap: TPRM data-source breadth vs dedicated TPRM suites | +| 7 | ~~No risk register / ERM primitive~~ — a lightweight `RiskRegister` model + risk-quantify surface **shipped** (v0.9.x–v0.10.x) | Honest framing: still NIST SP 800-30-shaped, not a Riskonnect-class ERM register | | 8 | **No reference customers / logo wall / ARR** | Pre-traction. v0.8 priority: get 5 named adopters (consultancies, FedRAMP CSPs, defense contractors) | | 9 | **No formalized SOC 2 / ISO 27001 of Evidentia itself** | OSS can't *be* SOC 2 compliant (it's not an entity). Mitigation: publish a formal threat model + supply-chain attestation roadmap; if a hosted-service offering ships, pursue SOC 2 Type II for that | -| 10 | **No IPO/exit narrative for vendor-longevity-paranoid buyers** | Mitigation: lean into "OSS = no acquisition risk; the code outlives any company"; cite GovReady-Q / Olive AI / Lacework valuation collapses as cautionary tales | +| 10 | **No IPO/exit narrative for vendor-longevity-paranoid buyers** | Mitigation: lean into "OSS = no acquisition risk; the code outlives any company"; cite GovReady-Q dormancy + Olive AI / Lacework valuation collapses as cautionary tales | --- @@ -857,7 +870,7 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a 1. **A maintained Python OSCAL library.** `oscal-pydantic` is 2+ years stale (last push 2024-04-06, 23 stars). Compliance-trestle is opinionated workflow software, not a library. Nothing else fills the `pip install evidentia-core; from evidentia_core import GapAnalyzer` shape. -2. **Bundled multi-framework crosswalk dataset.** NIST publishes 800-53 in OSCAL. CIS publishes Controls v8. ISO 27001 OSCAL ports are fragmented and unofficial. **Nobody bundles 88+ frameworks with vetted crosswalks** (Evidentia at v0.7.8; growing to 95+ in v0.7.9 with FFIEC + OCC + SR 11-7 additions). Evidentia could publish `evidentia-catalogs` as a separate Apache/CC0 repo and become the *de facto* multi-framework reference dataset, the way `oscal-content` is for NIST 800-53 alone. Includes the GovReady-Q vacated niche (210 stars; no meaningful work since 2024). +2. **Bundled multi-framework crosswalk dataset.** NIST publishes 800-53 in OSCAL. CIS publishes Controls v8. ISO 27001 OSCAL ports are fragmented and unofficial. **Nobody bundles 96 frameworks with vetted crosswalks**. Evidentia could publish `evidentia-catalogs` as a separate Apache/CC0 repo and become the *de facto* multi-framework reference dataset, the way `oscal-content` is for NIST 800-53 alone. Includes the niche GovReady-Q left dormant (210 stars; default-branch commits end 2023-06; last push 2024-12). 3. **Apache-2.0 alternative to AGPL CISO Assistant + Comp AI.** Many enterprises (defense contractors, SaaS vendors who would distribute the code, federal sub-primes) categorically refuse AGPL. There is no other permissively-licensed, library-first, OSCAL-native, multi-framework Python GRC toolkit with a cryptographically-signed evidence chain in active development today — the nearest exception at the platform level is theopenlane/core (Openlane): Apache-2.0, full-stack, Go (zero Python), very active, but its OSCAL support is schema-named fields only (no full OSCAL read/write) and it has no air-gap story. **Sharpened in v0.7.8**: CISO Assistant pulls 4,008 stars + daily commits + dual-license commercial paid SKUs; Comp AI raised a $2.6M pre-seed (announced 2025-08-01, co-led by OSS Capital + Grand Ventures) — the AGPL incumbents are well-staffed and ship daily. Evidentia's Apache 2.0 + Sigstore + SLSA L3 is the differentiable moat against them. @@ -896,7 +909,7 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a > Customer flight risk creates real opening for OSS that has no > acquisition exposure. -### 8.1 The four forcing-function dates (Q4 2026 convergence) +### 8.1 The forcing-function dates (a stacked multi-year mandate runway) | Date | Mandate | Implication for Evidentia | |---|---|---| @@ -904,7 +917,7 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a | **2026-05** target window | **HIPAA Security Rule final rule** (NPRM published 2025-01-06; OCR's regulatory agenda lists May 2026; ~4,700+ comments closed 2025-03-07; once final, 240 days to comply) | Significant uncertainty under new administration; bundled HIPAA Security catalog must be ready for revision | | **~~2026-08-02~~ → 2027-12-02** | **EU AI Act Annex III high-risk systems obligations DEFERRED to December 2, 2027** per the Digital Omnibus political agreement reached by the EU Council + Parliament on May 7, 2026 ([Council of EU press release](https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/); [IAPP coverage](https://iapp.org/news/a/ai-act-omnibus-what-just-happened-and-what-comes-next); [Hogan Lovells](https://www.hoganlovells.com/en/publications/eu-legislators-agree-to-delay-for-highrisk-ai-rules)). Article 50 transparency obligations apply from 2026-08-02 unchanged; only the Art 50(2) watermarking obligation gets a grace period to 2026-12-02 for existing systems under the May 2026 omnibus agreement. The Annex III deferral is per the 2026-05-07 political agreement (secondary-corroborated; formal adoption / OJ publication pending) but treated as the operative timeline by EU compliance counsel as of May 2026. | First likely AI Act fine pushed ≥16 months; ISO 42001 + EU AI Act crosswalks remain valuable but lose their Q3-Q4 2026 urgency; Evidentia's v0.9.3 EU AI Act tier classifier + v0.11+ ISO 42001 conformity-assessment-pack work retains relevance for the new Dec 2027 horizon | | **2026-09-11** | **EU CRA reporting obligations** — 24/72-hour incident + 14-day vulnerability reporting for any digital product placed on the EU market (CRA in force since 2024-12-10; full obligations including SBOMs apply 2027-12-11) | Hard demand spike for SBOM, vulnerability-tracking, incident-disclosure tooling; Evidentia's CycloneDX SBOM + collector pattern + audit-event stream is precisely positioned | -| **~~2026-09-30~~ → 2027-11-01** | **FedRAMP machine-readable packages** — NTC-0009 (2026-03-25) superseded RFC-0024's original 2026-09-30 deadline: comprehensive machine-readable (OSCAL) packages required by **2027-11-01** for Class D (High) only, with Classes A–C submitting semi-structured data by the same date (finalized in CR26; see §11.2.A). FedRAMP Ready still retires 2026-07-28 | OSCAL remains the load-bearing federal format, but on a later, narrower timeline; near-term urgency concentrates in Class D (High) CSPs; 90%+ of incumbents still don't natively emit OSCAL | +| **~~2026-09-30~~ → 2027-11-01** | **FedRAMP machine-readable packages** — NTC-0009 (2026-03-25) superseded RFC-0024's original 2026-09-30 deadline: comprehensive machine-readable packages (format-agnostic — NTC-0009 names no required format) required by **2027-11-01** for Class D (High) only, with Classes A–C submitting semi-structured data by the same date (finalized in CR26; see §11.2.A). FedRAMP Ready still retires 2026-07-28 | OSCAL remains a load-bearing federal rail (the Rev 5 / M-24-15 package path), with CR26's JSON schemas as the second machine-readable rail, on a later, narrower timeline; near-term urgency concentrates in Class D (High) CSPs; 90%+ of incumbents still don't natively emit OSCAL | | **2025-10-01** (already effective) | **Maryland MODPA** — strictest US privacy law (sale-of-sensitive-data prohibition + "reasonably necessary" data-minimization standard; $10K/violation, $25K/repeat penalties); effective 2025-10-01, with a carve-out exempting processing that occurred before 2026-04-01 | Bundled Tier-D state privacy laws now total 19 active in 2026; MODPA new high-water mark | | **2026-11-10** | **CMMC Phase 2** — Level 2 C3PAO third-party assessments mandatory for most DOD contractors handling CUI (Phase 1 self-assess became binding 2025-11-10) | **~80K DOD contractors need Level 2 certification (~220K total DIB) and 800-171 evidence collection**; assessment capacity: 103 authorized C3PAOs + 759 CCAs (Cyber AB, Mar 2026); Vanta/Drata at $30K+/yr unaffordable for most; Evidentia's free Apache 2.0 + air-gap support is the obvious fit | | **2026-11-10 to 2027-11-10** | DORA first full enforcement year — ESAs published list of **19 Critical ICT Third-Party Providers** (CTPPs) on 2025-11-18 (incl. AWS, Azure, GCP, Bloomberg, LSEG); 2026 Register of Information cycle uses 2025-12-31 reference data | EU CTPP oversight active; DORA-aware GRC tooling has captured customer base | @@ -940,7 +953,7 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a ### 8.4 Why this matters for Evidentia specifically -- **OSCAL-native, library-first** rides the FedRAMP machine-readable mandate (2027-11-01 for Class D High per NTC-0009; FedRAMP Ready retires 2026-07-28) at a moment when 90%+ of incumbents still don't natively emit OSCAL (Vanta added export Feb 2026 for FedRAMP 20x submissions only — the first of the Tier-1 SOC 2 automation platforms to do so; RegScale, Telos, Paramify, and C1Secure shipped commercial OSCAL tooling earlier) +- **OSCAL-native, library-first** rides the FedRAMP machine-readable mandates (CR26 JSON-MUST rules mandatory 2027-01-01; NTC-0009's format-agnostic comprehensive machine-readable requirement, 2027-11-01 for Class D High; FedRAMP Ready retires 2026-07-28) at a moment when 90%+ of incumbents still don't natively emit OSCAL (Vanta added export Feb 2026 for FedRAMP 20x submissions only — the first of the Tier-1 SOC 2 automation platforms to do so; RegScale, Telos, Paramify, and C1Secure shipped commercial OSCAL tooling earlier) - **Open-source + sovereign deploy** serves EU customers facing DORA enforcement + 19 CTPP oversight (who push toward EU-hosted or on-prem) and the ~80K DOD contractors (of a ~220K-company DIB) facing CMMC Level 2 certification under Phase 2 (who can't afford SaaS GRC at $30K+/yr) - **Sigstore + air-gap refusal** aligns with EO 14306 retained provisions and Defense Unicorns / Big Bang / Platform One patterns - **CycloneDX SBOM + collector pattern** aligns with CISA 2025 SBOM minimum elements update + FDA medical-device 524B + **EU CRA September 2026 reporting obligations** @@ -964,11 +977,11 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a > gravity wells; regulatory complexity demanding maintenance; AI hype > risk. None is fatal but each shapes positioning choices. -1. **Vanta + Drata + Optro + OneTrust have AI moats from scale.** Hundreds of integrations, customer corpora to fine-tune AI agents, 35+ pre-mapped frameworks. An OSS tool with 8 evidence-collection surfaces looks small by comparison. **Mitigation**: lean into OSCAL/sovereign/air-gap/cryptographic-evidence-attestation angles where SaaS can't follow; federate (let users plug Evidentia into existing SaaS GRC via the v0.7.8 Tableau / Power BI publish surfaces); position as the substrate that AI agents validate against (signed evidence, OSCAL artifacts, DFAH replay). +1. **Vanta + Drata + Optro + OneTrust have AI moats from scale.** Hundreds of integrations, customer corpora to fine-tune AI agents, 35+ pre-mapped frameworks. An OSS tool with 14 evidence collectors looks small by comparison. **Mitigation**: lean into OSCAL/sovereign/air-gap/cryptographic-evidence-attestation angles where SaaS can't follow; federate (let users plug Evidentia into existing SaaS GRC via the v0.7.8 Tableau / Power BI publish surfaces); position as the substrate that AI agents validate against (signed evidence, OSCAL artifacts, DFAH replay). 2. **NEW: AI feature commoditization (Q1 2026 universal launches).** Vanta, Drata, Optro, OneTrust, Hyperproof, Workiva, ServiceNow, Anecdotes all shipped agentic AI + MCP server + citation-grounded LLM output between Sept 2025 and April 2026. **The differentiation is collapsing rapidly** into eval rigor, audit-trail quality, and OSS license posture — not "we have AI." **Mitigation**: accelerate the v0.8.0 DFAH + PRT + MCP + plugin-contract differentiators before they become parity items. -3. **OSCAL adoption lags the mandate.** Zero of 100+ FedRAMP Rev5 authorizations in 2025 used OSCAL. The forcing function is real but vendors lag. **Vanta added OSCAL export Feb 2026 for FedRAMP 20x submissions only** — the first of the Tier-1 SOC 2 automation platforms to ship OSCAL export (commercial OSCAL tooling itself predates it: RegScale, Telos, Paramify, C1Secure). **Risk**: the market educates slowly; revenue follows mandate enforcement, not mandate publication. **Mitigation**: be patient; the September + November 2026 dates will pull the market forward; Evidentia is OSCAL-native and 18 months ahead of incumbent emit capability. +3. **OSCAL adoption lags the mandate.** Zero of 100+ FedRAMP Rev5 authorizations in 2025 used OSCAL. The forcing function is real but vendors lag. **Vanta added OSCAL export Feb 2026 for FedRAMP 20x submissions only** — the first of the Tier-1 SOC 2 automation platforms to ship OSCAL export (commercial OSCAL tooling itself predates it: RegScale, Telos, Paramify, C1Secure). **Risk**: the market educates slowly; revenue follows mandate enforcement, not mandate publication. **Mitigation**: be patient; the CR26 mandatory-for-all date (2027-01-01) and the NTC-0009 machine-readable deadline (2027-11-01) will pull the market forward; Evidentia is OSCAL-native today, ahead of most incumbents' native emit capability. 4. **NEW: Federal SBOM tailwind weakened (OMB M-26-05, January 2026).** [Davis Wright Tremaine analysis](https://www.dwt.com/blogs/privacy--security-law-blog/2026/02/omb-changes-course-on-software-security): M-26-05 rescinded M-22-18 + M-23-16, removing the mandatory standardized self-attestation form and shifting to risk-based agency discretion. SBOMs still recommended for cloud-runtime production environments. EO 14028 remains in force. **Mitigation**: EU CRA reporting obligations Sept 11 2026 more than offset globally; positioning toward EU-regulated digital-product manufacturers + US cloud-runtime SBOM (still mandated under M-26-05) carries the demand. @@ -980,7 +993,7 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a 8. **AI hype risk + Delve scandal cautionary baseline.** "Agentic GRC" is the hottest framing at every conference. An OSS tool that's "just a library" may look unfashionable to enterprise buyers chasing the AI checkbox. **March 2026 Delve scandal** (493/494 SOC 2 reports byte-identical; whistleblower disclosure DeepDelver; FTC + DOJ False Claims Act exposure) is the precise counter-narrative — "AI-powered" without rigor failed at scale. **Counter-positioning**: cite GRC 2020 March 2026 critique (most "agentic AI" features are AI-assisted form fills, not bounded-autonomy multi-step orchestrators); cite Magesh / Ho et al. (Stanford 2025) measured **17–33% hallucination rates** in commercial RAG-based legal research tools marketed as "hallucination-free"; cite PCAOB AS 1105 + Generative AI Spotlight as the regulator-recognized framework. Frame Evidentia's AI-optional posture as **deterministic-where-it-must-be, AI-where-it-helps**. Khatchadourian DFAH paper (March 2026, arXiv 2601.15322) provides the technical vocabulary. -9. **CMMC backlash possibility.** If Phase 2 (Nov 2026) gets delayed (it has been pushed before), the urgency tailwind softens. Mitigation: don't bet messaging exclusively on CMMC; diversify across the four forcing-function dates and the regulatory-vacuum opportunity (SR 11-7 → SR 26-02 model-risk gap). +9. **CMMC backlash possibility.** If Phase 2 (Nov 2026) gets delayed (it has been pushed before), the urgency tailwind softens. Mitigation: don't bet messaging exclusively on CMMC; diversify across the stacked forcing-function dates and the regulatory-vacuum opportunity (SR 11-7 → SR 26-02 model-risk gap). 10. **Federal procurement budget pressure.** 2025-2026 saw new administration's DOGE-driven cuts to consultants and contractors hitting major federal-services-sector stocks Feb 2025; FY26 continuing resolutions have squeezed civilian-agency + consultant compliance-tool spend ([Nextgov — agency consultant culling](https://www.nextgov.com/acquisition/2025/02/trump-administration-asks-agencies-cull-consultants/403345/)). Mitigation: free Apache 2.0 OSS is precisely the budget-pressure-friendly answer. @@ -1016,7 +1029,7 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a > *"Evidentia produces six attestable supply-chain credentials per release: PEP 740 attestations on every wheel, Sigstore-based signing, SLSA Level 3 build provenance, cosign-signed container images, CycloneDX 1.7 SBOMs, OpenSSF Scorecard 6.5+. OpenSSF Best Practices Silver MET, OSPS Baseline Maturity 2 MET. SOC 2 Type I examination in progress. Polycentric Labs LLC, Virginia."* -> *"Gap analysis in five interchange formats: OSCAL Assessment Results (1.2.1), SARIF 2.1.0, OCSF Compliance Finding (class_uid 2003), OCSF Detection Finding (class_uid 2004), CycloneDX 1.6 VEX. One toolkit, five SIEM / GRC platform / supply-chain ingestion paths. 95 bundled framework catalogs."* +> *"Gap analysis in five interchange formats: OSCAL Assessment Results (1.2.1), SARIF 2.1.0, OCSF Compliance Finding (class_uid 2003), OCSF Detection Finding (class_uid 2004), CycloneDX 1.6 VEX. One toolkit, five SIEM / GRC platform / supply-chain ingestion paths. 96 bundled framework catalogs."* > *"The puck is moving from PDFs of evidence toward signed JSON streams of evidence. Evidentia is sitting almost exactly where the regulatory and market vectors are converging in the next 12-18 months — and unlike SaaS GRC dashboards, every output artifact is independently verifiable against NIST SP 800-218 SSDF, NIST 800-53 SR/SA controls, and the CISA Secure-by-Design Pledge."* @@ -1024,9 +1037,9 @@ These are the gaps that will get a sales-eng on a Vanta/Drata bake-off to win, a ### 10.3 The supporting messaging hierarchy (no metaphors) -- **Primary**: *"Open-source (Apache 2.0) Python GRC toolkit by Polycentric Labs LLC (Virginia). 95 bundled framework catalogs. OSCAL 1.2.1-native. Six attestable supply-chain credentials per release (PEP 740 + Sigstore + SLSA L3 + cosign + CycloneDX 1.7 + OpenSSF Scorecard 6.5+) mapped to NIST SSDF + 800-53 + CISA SbD. Air-gap capable. AI-optional."* +- **Primary**: *"Open-source (Apache 2.0) Python GRC toolkit by Polycentric Labs LLC (Virginia). 96 bundled framework catalogs. OSCAL 1.2.1-native. Six attestable supply-chain credentials per release (PEP 740 + Sigstore + SLSA L3 + cosign + CycloneDX 1.7 + OpenSSF Scorecard 6.5+) mapped to NIST SSDF + 800-53 + CISA SbD. Air-gap capable. AI-optional."* - **Secondary**: *"Five gap-analysis interchange formats: OSCAL AR + SARIF + OCSF Compliance + OCSF Detection + CycloneDX VEX. 13-tool MCP server with CIMD scope + SignedToolOutput envelope. DFAH determinism + PRT policy-reasoning-traces eval rigor (first — and only known — OSS GRC-toolkit implementation of the Khatchadourian / Imperial 2026 academic frameworks)."* -- **Tertiary** (proof points): *"OpenSSF Best Practices Silver tier MET. OSPS Baseline Maturity 2 MET (~95% conforming, with explicit honest-gap declared for OSPS-AC-04.01/02 + OSPS-QA-07 + OSPS-GV-04 — the three solo-maintainer-structurally-unreachable controls). SOC 2 Type I examination in progress with Polycentric Labs LLC as the service organization, scope locked to development lifecycle + release pipeline + vendor management + internal IT controls. 17+ consecutive PROCEED-CLEAN releases of the v0.7.x → v0.10.x line. 3,440+ tests; mypy strict 0 issues across 278 source files; ruff clean."* +- **Tertiary** (proof points): *"OpenSSF Best Practices Silver tier MET. OSPS Baseline Maturity 2 MET (~95% conforming, with explicit honest-gap declared for OSPS-AC-04.01/02 + OSPS-QA-07 + OSPS-GV-04 — the three solo-maintainer-structurally-unreachable controls). SOC 2 Type I examination in progress with Polycentric Labs LLC as the service organization, scope locked to development lifecycle + release pipeline + vendor management + internal IT controls. 17+ consecutive PROCEED-CLEAN releases of the v0.7.x → v0.10.x line. 4,900+ tests; mypy strict 0 issues across 292 source files; ruff clean."* - **Honest disclaimer** (builds credibility): *"Not a Vanta replacement for SOC-2-only buyers who want a SaaS dashboard + auditor handoff bundled together. Evidentia is the OSS toolkit for security/compliance engineers who want every evidence artifact independently verifiable. OpenSSF Gold tier is structurally unreachable for a solo maintainer (Gold requires ≥2 unassociated significant contributors + bus factor 2+; no documented exception path); this gap is being resolved through the additional governance + segregation-of-duties controls implemented as part of the in-progress SOC 2 Type I program."* ### 10.4 The acceptance-gate paragraph (procurement-ready, Fortune 500 + federal-SI dual-audience) @@ -1134,7 +1147,7 @@ Three concrete uses: 3. **Pre-empt the FedRAMP CR26 OSCAL machine-readable narrative**: add a "FedRAMP CR26 + RFC-0024 readiness" callout to README + capability-matrix. The relevant dates per Stream 3 - NOTICE-0009 (March 25 2026): CR26 effective July 1 2026, mandatory + NOTICE-0009 (March 25 2026): CR26 launched June 25 2026 (SDR rules in force for 20x since July 4 2026), mandatory adoption Jan 1 2027; RFC-0024 deadline Nov 1 2027 + scope narrowed to Class D (High) only. diff --git a/docs/walkthrough-federal-si.md b/docs/walkthrough-federal-si.md index 930adf36..1a52c073 100644 --- a/docs/walkthrough-federal-si.md +++ b/docs/walkthrough-federal-si.md @@ -1,4 +1,4 @@ -# Federal-SI walk-through (v0.9.4 P3.1 + v0.9.5 P2.1/P2.2 refinement) +# Federal-SI walk-through > Operator persona: federal Systems Integrator (SI) procurement > officer at a CSP candidate. Walks the operator through a realistic @@ -7,15 +7,20 @@ > > Deferred since v0.9.0 §31.A. v0.9.4 P3.1 shipped the synthetic data > + the initial recipe. v0.9.5 P2.1 added AI-persona validation + -> v0.9.5 P2.2 added the POA&M → OSCAL emit step. Smoke-tested by -> `tests/integration/test_walkthrough_federal_si.py`. +> v0.9.5 P2.2 added the POA&M → OSCAL emit step; refreshed for the +> v0.11 federal wave (M-25-21 practice tracking, M-25-22 acquisition +> lifecycle, CR26 SDR/KSI emission — see "What's new at v0.11" below). +> Smoke-tested by `tests/integration/test_walkthrough_federal_si.py`. ## Trustworthiness of Evidentia itself (read first) Before sliding Evidentia into a federal authorization boundary, -verify the tool meets self-attestation expectations under EO 14028 -+ OMB M-22-18 + CISA Secure by Design Pledge + NIST SP 800-218 -SSDF practice PS.3.1 (cryptographic verification of integrity): +verify the tool meets software-supply-chain expectations under EO +14028 (as amended by OMB M-26-05, January 2026, which rescinded the +M-22-18 / M-23-16 self-attestation-form regime in favor of +risk-based agency discretion) + the CISA Secure by Design Pledge + +NIST SP 800-218 SSDF practice PS.3.1 (cryptographic verification of +integrity): ```bash # 1. PyPI PEP 740 attestations (signed by GitHub Actions OIDC @@ -36,18 +41,18 @@ cosign verify ghcr.io/polycentric-labs/evidentia:vX.Y.Z \ # your SBOM aggregator: Dependency-Track / Trivy / etc.). ``` -All three are required artifacts under FedRAMP RFC-0024 -(machine-readable authorization packages, **Nov 1 2027 initial- -compliance deadline** for Class D / High-impact CSPs per NOTICE- -0009 published March 25 2026 — the original Sept 30 2026 deadline -was superseded + scope was narrowed) and the CISA Secure Software -Self-Attestation Form. The companion **FedRAMP CR26** consolidated -rules (public preview May 4 2026; effective July 1 2026; -mandatory Jan 1 2027) replace the patchwork of post-FedRAMP- +All three artifacts serve the FedRAMP machine-readable direction +(NTC-0009, published March 25 2026: comprehensive machine-readable +authorization packages by **Nov 1 2027** for Class D / High-impact +CSPs — the original RFC-0024 Sept 30 2026 deadline was superseded + +scope was narrowed; agencies may still request SBOMs under +M-26-05's risk-based posture). The companion **FedRAMP CR26** consolidated +rules (public preview May 4 2026; launched June 25 2026 with the +SDR rules in force for 20x since July 4 2026; mandatory Jan 1 2027) replace the patchwork of post-FedRAMP- Authorization-Act memos with a declarative-rule format published in GitHub/FedRAMP/rules. Classes A/B/C only need "semi-structured text-based" data; Class D needs comprehensive machine-readable. -If the verify steps fail, file a v0.9.5 supply-chain incident +If the verify steps fail, file a supply-chain incident ticket against the repo — that's a stop-ship. ## Why this walk-through exists @@ -63,7 +68,7 @@ artifacts this persona produces: residual-risk justifications before re-authorization signature. 3. **FedRAMP PMO / DoD CCSP** — ingests machine-readable authorization packages per RFC-0024 (Class D / High-impact: - Nov 1 2027 initial-compliance deadline per NOTICE-0009 March 25 + Nov 1 2027 comprehensive machine-readable deadline per NOTICE-0009 March 25 2026; superseded the original Sept 30 2026 program-wide deadline); grades against the FedRAMP POA&M Template Completion Guide v3.0. @@ -80,15 +85,16 @@ shipping AI-powered systems into regulated environments: metric that matters at the AO desk. 2. **AI-governance attestation** for AI/ML systems the SI deploys to support federal contracts. The PRIMARY federal lens is - **OMB M-24-10** (Rights-Impacting / Safety-Impacting / Neither - categorization) plus **NIST AI RMF 1.0** (Govern / Map / - Measure / Manage functions). EU AI Act tier classification is - SECONDARY — useful when the same AI system also serves EU + **OMB M-25-21** (the single "high-impact AI" determination that + replaced M-24-10's Rights-Impacting / Safety-Impacting / Neither + categorization on 2025-04-03) plus **NIST AI RMF 1.0** (Govern / + Map / Measure / Manage functions). EU AI Act tier classification + is SECONDARY — useful when the same AI system also serves EU customers, but not the headline lens at a federal AO desk. Both surfaces have hard deadlines (monthly POA&M attestation; NIST AI RMF / AI Executive Order obligations on federal AI use -cases; **FedRAMP CR26 effective July 1 2026 / mandatory Jan 1 +cases; **FedRAMP CR26 live since June 25 2026 / mandatory Jan 1 2027** — the most consequential FedRAMP structural change in a decade; **FedRAMP RFC-0024 machine-readable Class D / High-impact authorization packages: Nov 1 2027** per NOTICE-0009 March 25 @@ -123,8 +129,9 @@ evidentia catalog list | grep -E "(nist|fedramp|nist-ai-rmf|eu-ai-act)" **Expected**: at minimum the bundled NIST 800-53 Rev 5 (baselines Low/Moderate/High; 156/323/443 controls respectively after RFC- -0027/0030), the FedRAMP overlay, NIST AI RMF 1.0, and EU AI Act -statutory excerpts. Evidentia ships 89 framework catalogs by +0027/0030), the FedRAMP overlay (Rev 5 baselines + the CR26 +`fedramp-ksi-2026` KSI catalog), NIST AI RMF 1.0, and EU AI Act +statutory excerpts. Evidentia ships 96 framework catalogs by default; this filter narrows to the ones the SI workflow exercises. @@ -140,7 +147,7 @@ narrowed; Classes A/B/C only need semi-structured text-based data). ```bash evidentia conmon check \ - --last-completed-file tests/data/walkthrough-federal-si/state.yaml \ + --state-file tests/data/walkthrough-federal-si/state.yaml \ --today 2026-05-18 ``` @@ -175,10 +182,11 @@ companion `GET /api/conmon/daemon-status` health-check endpoint, and the v0.9.5 P2.3 `--history-file` for the `GET /api/conmon/daemon-history` flap-detection endpoint. -> **CLI flag naming note**: `conmon check` uses `--last-completed- -> file`; `conmon health` + `conmon watch` use `--state-file` for -> the same on-disk YAML schema. This is a historic inconsistency -> tracked for a v0.9.6 normalization pass. +> **CLI flag naming note**: `conmon check` historically used +> `--last-completed-file`; v0.9.6 normalized every conmon verb to +> `--state-file`. The legacy flag still works but is deprecated +> (removal targeted at v1.0.0 per +> [`deprecation-calendar.md`](deprecation-calendar.md)). ## Step 3 — Compute framework health @@ -305,7 +313,8 @@ ownership + classification. Federal-SI persona note: promoting a federal-AI system from pilot → production *within* an authorization boundary typically triggers a **FedRAMP Significant Change Request (SCR)** plus an -**OMB M-24-10 AI Use Case Inventory update**. Evidentia's +**AI Use Case Inventory update** (the inventory obligation carried +forward under OMB M-25-21). Evidentia's `ai-gov update` fires the `AI_SYSTEM_UPDATED` lifecycle audit event for downstream automation, but does **NOT** auto-emit the SCR Form (the FedRAMP-PMO-required template). Operators produce @@ -370,8 +379,9 @@ evidentia oscal verify /tmp/walkthrough-poam.oscal.json document with SHA-256 back-matter integrity references to every embedded POA&M record. This document is what the SI submits to the 3PAO at annual assessment and to the FedRAMP PMO at the monthly -POA&M cadence (post-RFC-0024 deadline, this is the machine- -readable submission format). +POA&M cadence (OSCAL POA&M is the established Rev 5 machine-readable +package rail; NTC-0009's 2027-11-01 comprehensive machine-readable +requirement is format-agnostic). Federal-SI persona note: the FedRAMP POA&M Template Completion Guide v3.0 prescribes specific column semantics (Severity, @@ -379,18 +389,71 @@ Detection Source, POA&M Items Open, Original Detection Date, Risk Adjustment, Deviation Request, etc.) that Evidentia's OSCAL emit preserves via OSCAL's prop+annotation mechanism. The v3.0 template's MS Excel form remains the FedRAMP PMO ingest -channel; FedRAMP CR26 + RFC-0024 (Class D / High-impact: Nov 1 -2027 initial-compliance; Sept 30 2027 full final) transitions the +channel; FedRAMP CR26 + NTC-0009 (Class D / High-impact: Nov 1 +2027 comprehensive machine-readable) transitions the PMO to machine-readable JSON. OSCAL 1.2.1 emit interoperates with both. (Evidentia has emitted OSCAL 1.2.1 since v0.9.6, aligned with the upstream compliance-trestle library.) +## What's new at v0.11 — the federal wave + +Three v0.11 surfaces extend the workflow above; they are additive +(steps 1–8 are unchanged) and map directly onto this persona: + +1. **M-25-21 minimum-practice tracking** (extends Steps 4–5). Once a + system carries a high-impact determination, record per-practice + status for the memo's seven §4(b) minimum practices — including + CAIO waiver objects with the annual re-certification and 30-day + OMB-report clocks: + + ```bash + EVIDENTIA_AI_REGISTRY_DIR=/tmp/walkthrough-federal-si-registry \ + evidentia ai-gov set-practice \ + --practice pre_deployment_testing --status implemented + ``` + +2. **M-25-22 acquisition lifecycle** (extends Step 7 — this persona + IS the procurement officer M-25-22 addresses). Track an AI + acquisition through the memo's six §4 lifecycle phases + (`identification_of_requirements` → `contract_closeout`), tied to + the §4(a) initial high-impact determination: + + ```bash + EVIDENTIA_AI_ACQUISITION_DIR=/tmp/walkthrough-federal-si-acq \ + evidentia ai-gov acquisition register "Resume-screener SaaS" \ + --solicitation-ref RFP-2026-001 \ + --likely-high-impact high_impact \ + --link-system + EVIDENTIA_AI_ACQUISITION_DIR=/tmp/walkthrough-federal-si-acq \ + evidentia ai-gov acquisition set-phase \ + --phase market_research_and_planning --status in_progress + ``` + +3. **FedRAMP CR26 SDR / KSI emission** (extends Steps 2–3 for the + 20x path). Emit the Security Decision Record + `keySecurityIndicators` block (10 families / 46 KSIs) conformant + to the official 2026-06-24 schemas — the machine-readable + companion to the OSCAL POA&M heartbeat in Step 8. Like Step 8, + this is operator-driven: `--status-file` takes your own + per-KSI status YAML (`KsiStatusDocument` schema): + + ```bash + evidentia conmon ksi \ + --status-file ./my-ksi-status.yaml \ + --state-file tests/data/walkthrough-federal-si/state.yaml \ + --out /tmp/walkthrough-sdr-skeleton.json + ``` + + See [`fedramp-ksi.md`](fedramp-ksi.md) for the status-file + schema, schema provenance, vendored-pin policy, and the weekly + drift sentinel. + ## What this walk-through validates | Phase | Capability tested | Federal-SI consumer | |---|---|---| | 0 | Supply-chain trustworthiness (PEP 740 + cosign + SBOM) | Procurement officer self-attestation | -| 1 | Catalog discovery — 89 bundled frameworks load | All | +| 1 | Catalog discovery — 96 bundled frameworks load | All | | 2 | CONMON cycle classification — overdue/due_soon bucketing | SI compliance ops; internal SIEM | | 3 | Framework health scoring — JSON output for SIEM ingest | SI compliance ops only (NOT PMO-grade) | | 4 | EU AI Act tier classification — Annex III + risk attributes | Procurement evaluation; pairs with OMB M-25-21 high-impact | @@ -399,7 +462,7 @@ the upstream compliance-trestle library.) | 7 | Lifecycle CLI verbs (v0.9.4 P2.3) — update + retire firing audit events | SI compliance ops; pairs with SCR Form out-of-band | | 8 | POA&M emit + OSCAL 1.2.1 plan-of-action-and-milestones | 3PAO annual assessment; FedRAMP PMO monthly POA&M; RFC-0024 | -If any step diverges from "expected", file a v0.9.6 bug ticket +If any step diverges from "expected", file a bug ticket with the actual output. The integration test `tests/integration/test_walkthrough_federal_si.py` runs steps 2-4 + 6 in CI to catch regressions in the smoke surface; the full @@ -417,7 +480,8 @@ addressed: - **EU AI Act-first framing**: the v0.9.4 doc led with EU AI Act tier classification, which is the wrong primary lens for a federal-SI persona. v0.9.5 reframed with OMB M-24-10 + - NIST AI RMF as primary, EU AI Act as secondary. + NIST AI RMF as primary, EU AI Act as secondary (the primary lens + subsequently migrated to M-25-21 at v0.10.12, reflected above). - **CLI flag bug in Step 2**: v0.9.4 doc referenced `--state-file` for `conmon check` (which actually uses `--last-completed-file`). Fixed in v0.9.5. @@ -432,19 +496,20 @@ addressed: - **Health score conflation**: the 0.857 metric is internal dashboard, not PMO-grade. v0.9.5 added the explicit caveat. -Limitations still open for v0.9.6+: +Limitations still open (resolved items pruned — FIPS 199 became a +first-class registry field at v0.9.6 via `categorize-fips`; the +conmon flag naming was normalized to `--state-file` at v0.9.6): -- **FIPS 199 categorization on AI registry**: tracked as a - first-class field; today carried in the `owner` sidecar or via - the v0.9.5 Phase 3 custom fields. - **SCR Form auto-emit**: Evidentia fires the lifecycle audit event but does not generate the FedRAMP-PMO SCR template. Operator produces SCR out-of-band. -- **CLI flag naming consistency**: `conmon check` vs `conmon - health` use different flag names for the same on-disk schema. +- **GUI coverage of the v0.11 federal verbs**: `ai-gov + set-practice` and the `ai-gov acquisition` verbs are CLI/API-only + pending the next console pass (see + [`parity-coverage.md`](parity-coverage.md)). - **Real federal-SI operator review**: the v0.9.5 validation was AI-persona-driven; a real domain-expert review remains the - highest-value v0.9.6 follow-up. + highest-value follow-up. ## See also @@ -458,6 +523,11 @@ Limitations still open for v0.9.6+: Governance, and Public Trust (2025-04-03; rescinded + replaced OMB M-24-10's rights/safety-impacting model with a single "high-impact AI" category) +- OMB M-25-22 — Driving Efficient Acquisition of Artificial + Intelligence in Government (2025-04-03; replaced M-24-18; the six + §4 lifecycle phases Evidentia's acquisition surface models) +- `docs/fedramp-ksi.md` — the CR26 SDR / KSI emitter (schema + provenance + drift sentinel) - NIST AI RMF 1.0 — AI risk management framework - NIST SP 800-218 SSDF — Secure Software Development Framework - CISA Secure by Design Pledge — voluntary commitment to diff --git a/docs/wiki/2-guides/ai-governance.md b/docs/wiki/2-guides/ai-governance.md index 3c6b4bd5..b1319ab1 100644 --- a/docs/wiki/2-guides/ai-governance.md +++ b/docs/wiki/2-guides/ai-governance.md @@ -486,8 +486,13 @@ Valid `--determination` values: `high_impact` / `not_high_impact` / > risk-management practices (pre-deployment testing; AI impact assessment; > ongoing monitoring + adverse-impact detection; human training/competency; > enhanced human oversight/intervention; consistent remedies & appeals; -> end-user/public feedback). Evidentia records the determination for downstream -> OSCAL emit and reporting; per-practice tracking is a planned follow-up. +> end-user/public feedback). As of v0.11, record per-practice status with +> `evidentia ai-gov set-practice --practice --status ` +> — including §4(a)(ii) CAIO waivers with their annual re-certification and +> 30-day OMB-report clocks (`--waiver-*` flags). The companion M-25-22 +> acquisition surface (`evidentia ai-gov acquisition +> register|list|show|set-phase`) tracks AI procurements through the memo's +> six §4 lifecycle phases. > **Legacy / rescinded memo.** OMB M-24-10 was rescinded on 2025-04-03 by > M-25-21. The legacy `set-omb-impact --category